xbbg 1.2.2

Fixed

• Bloomberg request error handling and datetime semantics (#328, #329, #330): Reference-data wide/semi-long requests and historical-data requests now surface all-security securityError responses as BlpError::RequestFailure instead of returning successful empty batches. The Python terminal wrapper now handles pyarrow.RecordBatch returns via pyarrow.Table.from_batches() rather than assuming a non-existent to_table() method. Bloomberg datetime parsing now rejects malformed T datetime strings and SDK-invalid timezone offsets, and timestamp conversion honors Bloomberg's OFFSET part when converting to UTC epoch values.

• Python and JavaScript binding edge cases: xbbg.set_sdk_path() now immediately prepares native SDK loading and retains Windows DLL search handles, explicit but unimplemented Python dataframe backends now raise NotImplementedError instead of falling through to Narwhals/native output, turnover helpers reject malformed explicit dates before applying defaults, NAPI Arrow zero-copy length metadata is checked before narrowing to JavaScript-visible u32, and JS zero-copy Arrow construction validates buffer lengths before creating typed views.

• active_futures / activeFutures generic futures mapping (#327): The shared Rust recipe now uses Bloomberg's historical FUT_CUR_GEN_TICKER mapping as the primary source for past/current dates, normalizing returned roots such as UXK6 to the input asset suffix (UXK6 Index) across Python and JavaScript bindings. Future dates or missing mappings fall back to the existing generated-candidate maturity/volume logic. The recipe also accepts typed Arrow value columns (Float64 VOLUME, Date32 LAST_TRADEABLE_DT) instead of requiring Utf8, fixing the monthly/default-frequency failure for UX1 Index.

Full Changelog: v1.2.1...v1.2.2